Thomas House

That group in UKHSA is brilliant, btw.

When not being snarky, this is the right take. The "Lizzy line" is a good thing and makes the case for infrastructure spend elsewhere. Although the snark is understandable in such a capital-city-centred country.

Of course, in living memory Universities were publicly funded rather than via student fees and many countries still have that model.

> "Godlike intelligence trying to explain itself to we mere mortals".

I'm trying to wean myself off use of "AI" - what these models are is general functional learners optimised for prediction, and the explainable bit is trying to get a lower dimensional representation of the learned functional behaviour. So, not as snappy, but it avoids the implications of >

(like I say, not trying to start an argument and not an inconsistent point with your post)

Goodness there's arguments I really don't want to have online, but it's incredibly important to be clear that the Catholic civil rights movement in NI in the 1960s wasn't paramilitary.

There's an extension pack for Taco Cat Goat Cheese Pizza. No recommendation implied, just putting that fact out there.

I came to prefer "immunity gap" but yes, it's just a fact that infection rather than vaccination is what provides some of the immunity that keeps current numbers of infections relatively low.

There's always the issue though that the data need some adjustment - you can't get away from modelling that! But, one can try to use simpler models that make fewer assumptions. In a way there was more focus on doubling and having times because they delivered what policy needed at any time.

I can't claim to understand that either - except that for quite good reasons the ONS wants its outputs to be very tightly quality controlled and to leave higher uncertainty analysis to researchers.

Real-time modelling of the SARS-CoV-2 pandemic in England 2020–2023: a challenging data integration Abstract. A central pillar of the UK’s response to the SARS-CoV-2 pandemic was the provision of up-to-the moment nowcasts and short-term projections to mon

Here's one I had a minor involvement in - although with more focus on hospitalisation (Fig 3) than death. I think their real-time model is basically consistent with the whole pandemic but doesn't focus on one specific outcome.

academic.oup.com/jrsssa/advan...

I think many research papers aim for IFR by age, i.e. something more complex. It's implicitly there in e.g. the public Cambridge modelling that calibrates carefully to death data and ONS community infection.

There's potentially loads of well paid, safe, unionised industrial jobs retooling the economy for closed loop manufacturing, zero carbon etc.

People tend to write wrappers for the really good Fortran implementations of classic algorithms like L-BFGS-B widely used now in machine learning rather than reimplement - sometimes there's just almost no room for improvement.

Palestine Action protesters arrested by police at London demo The Metropolitan Police said the number of arrests was the largest made by the force on a single day in the last decade.

When I expressed free-speech doubts about the UK's 2000 Terrorism Act, people often said, "Of course it isn't going to impact on peaceful protest". Now people often say, "What do the peaceful protesters expect, they're breaking an established law?"

www.bbc.co.uk/news/article...

Chinese engineer working there? It seems like the system would be more robust to printing errors / poor resolution.

So, so much of the problem with "AI" discourse comes from viewing the new consumer products as anything like intelligences. Large Language Models is a completely fair and descriptive term - most important is the last term. These are *models* of natural language, with parameters that are >

He gives more those skeleton enemies you can't ever quite get rid of they keep reforming from piles of bones.

Jobs at Warwick Mathematics - absolutely great department and colleagues, somewhere you can very much ignore the "Beyond" silliness people have been teasing here and get on with good work.

warwick-careers.tal.net/vx/appcentre...

warwick-careers.tal.net/vx/appcentre...

Conceptual James still seems to be the personification of "execrable" though.

Actually sort of this - these tools are models of language, why should they be able to do maths or count letters?

bsky.app/profile/tah-...

Guy Gives Himself 19th Century Psychiatric Illness After Consulting With ChatGPT "For 3 months, he had replaced sodium chloride with sodium bromide obtained from the internet after consultation with ChatGPT."

PS, in the same way you shouldn't buy stock prices on the basis of an epidemic model, you shouldn't take medical advice from a language model. Remember, LLM. It's in the name.

www.404media.co/guy-gives-hi...

> But, in a way, that's also why it's such a shame that the hype obscures the fact that having better and new models of natural language is quite cool and useful! Just like when we get new models of all sorts of other phenomena.

YouTube video by Alberta Tech Why can't AI figure out this math problem? #ai

> What obscures this is the "Large" part - the amounts of data and number of parameters are so huge that we can't get any very good intuition for what is going on. I heartily chuckled at this video satirising the conclusions that can lead to. >

youtube.com/shorts/pAq0k...

> they can also be misspecified, there can be problems with over and under fitting, biases in data and issues if the target for prediction is outside the dataset. Our brains aren't models, at least not in the way LLMs are, they're a level up in that they can create and understand models. >

> fitted to data, just like we have models of heart attack risk, epidemics, stock prices, structural integrity of bridges etc. And so they exhibit very similar behaviour to all other models: at their best, they automate things that can take humans a long time to do and augment performance. But, >

So, so much of the problem with "AI" discourse comes from viewing the new consumer products as anything like intelligences. Large Language Models is a completely fair and descriptive term - most important is the last term. These are *models* of natural language, with parameters that are >

🚨 "Let's agree to disagree" is destroying relationships.

Communications professor Lisa Pavia-Higel from Missouri University of Science and Technology explains why this phrase hurts and what you can do instead.